How many degrees of freedom do we need to train deep networks: a loss landscape perspective.

Brett W. Larsen, Stanislav Fort, Nic Becker, Surya Ganguli

Browse the full ICLR paper archive.